A million Swedish men say smart people are nicer. Not so fast.
July 3, 2026 · Keith Stanovich, How to Think Straight About Psychology~6 min read
You know the feeling where a headline arrives with its homework already done? This week it was smart people are kinder — proven, apparently, by 1.2 million Swedes. And the number is what does the persuading. A dozen students, you'd shrug off. But a million-plus? That feels less like a study and more like a census of the truth. Which is exactly the trick I want to slow down on, because the size of a sample and the strength of its logic are two completely different things, and one is quietly borrowing the other's authority.
What the study actually did
The work is real and, on its own terms, rather impressive. Mikael Elinder and Oscar Erixson at Uppsala published it in The Economic Journal — an inquiry into intelligence and prosocial behaviour, drawn from Swedish population registers. The spine of it: roughly 1.2 million men, born between 1951 and 1979, who sat a mandatory cognitive test (logical, verbal, spatial, technical) at military enlistment in their late teens. Decades later, the researchers linked those teenage scores to what those same men did as adults, recorded in government registers. The associations are consistent and, in this sample, sizeable: a one-standard-deviation bump in test score came with roughly a 40% higher likelihood of making a charitable donation, about 31% higher odds of voting, and around 14% higher odds of owning an eco-friendly — electric or ethanol — car. Three different flavours of "good citizen," all leaning the same way. It is a genuinely careful piece of work, which, if we're being honest about it, is precisely what makes the headline on top of it so slippery.
Why "a million" feels like proof (and isn't)
Here's the thing about a giant sample: it buys you one specific thing, and it isn't the thing you feel it buys you. A huge N narrows your confidence interval — it makes the estimate of the association tight and stable, so you can be very sure the correlation is really there and not a fluke of small numbers. That's worth having. But precision about whether two things travel together tells you nothing about why. Keith Stanovich, in How to Think Straight About Psychology, keeps returning to this one unglamorous line in the sand: correlation is not causation, and no amount of data volume moves that line. If the link between test scores and giving is being driven by some third thing lurking underneath both, then measuring a million people instead of a thousand doesn't dissolve that third thing — it just measures the confounded relationship with heroic accuracy. You end up extremely confident about a claim that was never the claim you wanted.
A 2025 study in The Economic Journal (Elinder & Erixson) linked enlistment-test scores to giving, voting and eco-car ownership across ~1.2 million Swedish men. A one-SD higher score went with ~40% higher odds of donating, ~31% of voting, ~14% of owning an eco car. But education, income, family background and health all push on both ends, and a huge sample sharpens the estimate without removing the confound. Framework: Keith Stanovich, How to Think Straight About Psychology. Correlational, not causal; popular-science interpretation, not professional advice.
The third variables practically queue up
And in this case they don't even hide. Ask what else, besides some pure quality called "intelligence," might raise a teenager's test score. Education, for a start — more schooling, better schooling, a home full of books. Family income and background. Health, both mental and physical. Now ask what else predicts whether an adult donates to charity, turns up to vote, or can afford an electric car. Education. Income. Health. It's the same list, more or less — the exact soup of advantages that makes someone both test well at eighteen and vote and give and drive electric at forty. A kid raised with money, stability and good schooling scores higher on the enlistment test and, decades on, has the spare income to donate and the means to buy the eco car — with intelligence doing far less of the work than the headline needs it to. To their credit, the authors know this and reach for twin comparisons to hold some of it steady, which helps; but "helps" is not "settles," and reading registry data means accepting that a great many of these advantages are simply tangled together in ways no register can fully pull apart.
A test is not the whole of a mind
There's a quieter leap worth flagging too, the one that happens before any of the statistics. A cognitive test taken at a military recruitment desk — logical, verbal, spatial, technical — is a real and useful measure of certain abilities. It is not the same thing as "intelligence" in the full, human sense we mean when we call someone bright, and it is a very long way indeed from "goodness." When the coverage quietly slides from "scored well on an enlistment test" to "is more intelligent" to "is more prosocial" to "is a better person," each step sounds small and the sum is enormous. And notice who's even in the room: 1.2 million men, plus around 3,000 women who volunteered to enlist. Volunteering makes that group of women self-selected — not a random cross-section, but a particular sort of person who chooses the military — so whatever holds for them may not hold for women generally. Add that these are Swedish men of a specific set of birth years, a particular country and era, and the tidy universal claim starts to look like what it is: a precise finding about a particular slice of the world.
A million data points can tell you a correlation is real. They cannot tell you it's a cause.
Size buys you precision about the link — never permission to draw the arrow.
What this means for you
None of this means the finding is worthless, and I'd gently steer you away from the equal and opposite overreaction — deciding it's "just correlation" and therefore nothing. It's a real pattern in an enormous, high-quality dataset, and that's worth knowing. The useful move is smaller and calmer than either the celebration or the sneer. Next time a headline waves a giant number at you — a million people, a hundred studies, a huge dataset — treat the size as a claim about precision, not about cause, and ask Stanovich's plain question underneath it: what else, sitting under both of these things, could be driving them at once? If you can name three candidates before you've finished your coffee (here: schooling, money, health), the straight arrow in the headline was never the only story — just the flattering one. Which, honestly, is usually the tell.
Framework from Keith Stanovich's How to Think Straight About Psychology (斯坦诺维奇《对伪心理学说不》) — correlation is not causation, the third-variable / confound problem, why a large sample buys precision but not causal license, self-selection, and the limits of a single construct. News peg: Mikael Elinder & Oscar Erixson, "An Inquiry into the Relationship between Intelligence and Prosocial Behaviour: Evidence from Swedish Population Registers," The Economic Journal (2025); ~1.2 million Swedish men born 1951–1979 (plus ~3,000 women who volunteered to enlist), enlistment cognitive tests linked to registry records; a one-SD higher score associated with ~40% higher odds of donating, ~31% of voting, ~14% of owning an eco car; covered by PsyPost, July 2026. Key caveat: these associations are correlational, not causal, and this is a popular-science interpretation, not professional psychological advice.
框架取自基思·斯坦诺维奇《对伪心理学说不》——相关不是因果、第三变量/混杂问题、为什么大样本买来的是精确而非因果的许可、自我选择,以及单一测量的局限。新闻由头:Mikael Elinder 与 Oscar Erixson,《An Inquiry into the Relationship between Intelligence and Prosocial Behaviour: Evidence from Swedish Population Registers》,发表于《The Economic Journal》(2025);约 120 万名 1951–1979 年出生的瑞典男性(外加约 3000 名自愿入伍的女性),征兵认知测试与登记册记录对接;测试分每高 1 个标准差,捐款概率约高 40%、投票约高 31%、开环保车约高 14%;经 PsyPost 于 2026 年 7 月报道。关键限定:这些关联是相关性、非因果;本文为科普解读,非专业心理建议。
研究は本物で、その土俵でいえば、なかなか立派だ。ウプサラ大学の Mikael Elinder と Oscar Erixson が、『The Economic Journal』に載せた——知能と向社会的行動をめぐる探究で、スウェーデンの人口登録簿から引いている。背骨はこう。約120万人の男性、1951年から1979年の生まれで、十代の終わりに徴兵で強制の認知テスト(論理・言語・空間・技術)を受けている。数十年後、研究者はその十代の点数を、同じ男性たちが大人になってから政府の登録簿に記録された行動と結びつけた。関連は一貫していて、この標本ではそこそこ大きい。点数が1標準偏差高いと、寄付をする確率が約40%、投票する確率が約31%、環境にやさしい(電気かエタノールの)車を持つ確率が約14%高い。三つの違う味の「良き市民」が、そろって同じ方へ傾く。かく言う私も感心したけど、これはちゃんとした仕事だ。そして、ちゃんとしているからこそ、その上に乗った見出しが、やたらと滑りやすい。
枠組みはキース・スタノヴィッチ『心理学をまじめに考える方法』より——相関は因果ではない、第三変数/交絡の問題、大標本が買うのは精確さで因果の許可ではないこと、自己選択、そして単一の測定の限界。ニュースの起点:Mikael Elinder と Oscar Erixson「An Inquiry into the Relationship between Intelligence and Prosocial Behaviour: Evidence from Swedish Population Registers」、『The Economic Journal』(2025)。約120万人の1951–1979年生まれのスウェーデン男性(それに志願入隊した約3,000人の女性)、徴兵の認知テストを登録簿の記録と結合。点数が1標準偏差高いと、寄付の確率が約40%、投票が約31%、エコカー所有が約14%高い。PsyPost が2026年7月に報じた。主な但し書き:これらの関連は相関的で因果ではなく、本稿は科学解説であって、専門的な心理の助言ではない。